AI Hiring Needs Judgment, Not Just Code

A prompt library does not tell a charity who may sign an eligibility note. It does not tell a manufacturer when a generated checklist is wrong. The next hiring question is less spectacular than a model-building vacancy, and a coding test will not settle it.

That is an editorial claim. The new numbers come from a recruitment firm, and they need a denominator before anyone calls them a shortage.

The 61% is a comparison, not a surplus of coders

On 29 October 2026 Randstad published Tech talent in transition: building the AI-ready workforce, with a report overview and a press release. The primary research was done in the first half of August 2026. It was 1,100 interviews, 100 in each of 11 markets: Australia, Belgium, France, Germany, India, Italy, Japan, the Netherlands, Spain, the United Kingdom and the United States. Respondents were senior business and technology decision-makers responsible for workforce planning and digital talent strategy. Evalueserve was used for sampling. The pages do not publish a margin of error or the weights.¹²

The secondary analysis, compiled by Randstad Market Intelligence in August 2026, uses over 20 million job postings and approximately 100 million candidate profiles. A posting is not a completed hire. A profile is not a census of people who can do the work.²

61% is the figure that will travel. The pages do not use one sentence for it.

This article treats those lines as one survey result described four ways. It does not upgrade them into proof that the courses, the tests or the job titles were the same, or that coders are now easy to find.¹²

Technical demand is not collapsing beside that result. The press headline says entry-level tech jobs requiring AI skills are up fivefold since 2025. The body of the same release says postings requiring AI capabilities grew more than fivefold over the past year. The overview says entry-level postings requiring AI skills grew more than fivefold since 2025. Those are not one denominator. They do not show that every junior coding role became five times harder to fill.¹²

86% of the surveyed organizations are expanding technology headcount. The press headline also says 86% of employers are growing tech teams. That is this panel of senior decision-makers, not 86% of employers worldwide, and not a rise in total employment. Adding specialists can sit next to cuts elsewhere.¹²

The governance comparison is the one most often turned into a vacancy count. The press body says demand for risk governance doubled (+108%) while talent supply grew by 69%. The overview FAQ says demand for risk-governance roles doubled (+108%). The two percentages come from different kinds of data, postings on one side and profiles on the other, with definitions the public pages do not fully open. 108 minus 69 is not a count of empty seats.¹²

Randstad sells talent, including flexible staffing. The press calls flexible talent its respondents' first-choice answer: 63% name contingent, freelance, contract and fractional models to close AI skills gaps. 92% use digital talent marketplaces, and 42% use them as the primary channel. A vendor with a commercial interest in that answer can still publish a careful survey. The interest is a reason to keep an independent frame, not a reason to discard the pages.²

Sander van 't Noordende, Randstad's CEO, is quoted as saying employers need people who can put AI to work from day one and, increasingly, talent with the judgment, communication and governance skills to make sure it delivers, and that this combination is now the hardest aspect to hire for. That is the company's reading of its own study.²

Communication is part of the control system

Consider a customer-service team that adds an assistant to draft replies. Someone has to decide which requests it may answer, which data it may see, and when a person takes over. Someone has to explain to colleagues why a fast reply can still be the wrong one. Someone has to read the complaints and change the rule. None of that is solved by model fluency alone. In that workflow, communication is part of the control system.

The level of skill should follow the cost of an error. A person who uses a tool to tidy an internal note does not need the same oversight as a manager who lets an automated score shape a hiring decision, or a service that lets a summary affect eligibility. That distinction is an editorial one. A public-sector brief gives it a published shape.

On 19 January 2026 the OECD published Building an AI-ready public workforce. It is guidance for public administrations, built from the Alan Turing Institute's AI Skills for Business Competency Framework (2024) and the OECD Framework for Digital Talent and Skills in the Public Sector (2021). It names three groups. General employees need to use tools effectively and responsibly: basic digital skills, awareness of potential, risks and ethical use, data-protection awareness, and critical thinking and independent judgement. Leaders need a strategic view of risks, ethics and regulation, plus collaboration, communication, stakeholder management and change management. Digital and data professionals need the technical skill to support development, implementation and maintenance, including the ability to mitigate risk, and they still need to work across disciplines. One generic course for all three groups is a weak design. The brief does not test that design in a private charity or a factory.³

The OECD policy brief AI and skills: What we know so far, 5 June 2026, makes a related point about economies, not about one vacancy. It says one of the decisive factors in whether economies benefit from this shift may not be the technology itself, but the skills of the people using it. Less than 1% of workers need advanced AI skills such as programming or model development, a figure taken from Green and Lamby (2023). Most people need digital skills and the ability to use, analyse and interpret data, plus managerial and human skills such as problem solving, creativity and innovation. The brief says its most recent data date from late 2024. An older finding is not a 2026 hiring audit.⁴

The same brief refuses a simple story in which human skills only rise. In Germany, France, Italy and Spain, managers using algorithmic management tools are more likely to say those tools reduce the need for empathy (20%) than increase it (12%), from Milanez, Lemmens and Ruggiu (2025). The OECD says it is too early to draw firm conclusions. A hiring plan that treats "communication" as an automatic premium, without asking what the tool is doing to the work, is ahead of this evidence.⁴

The International Labour Organization's publication page of 13 August 2026, for a joint report with the European Training Foundation, Cedefop, Eurofound, the European Commission and UNESCO, says AI adoption often increases the need for higher-order cognitive and socioemotional skills, and for general digital and data science skills. Jobs that develop and maintain AI systems are described as currently small and niche, and as growing rapidly. The full PDF, 10.78 MB, was not re-read page by page.⁵

For the separate question of who received practice rather than a login, see why AI training reaches bosses more often than beginners. For checking what a system produced, see why human review becomes the bottleneck once agents are fast. For roles built around testing and governing systems, see who the oversight jobs are actually for.

A hypothetical Tuesday, not a case study

Imagine a 45-person social-services organization using a tool to summarize case notes. The summaries are shorter. A supervisor then sees that one of them drops a qualification that changes eligibility. Nobody meant to mislead a client. The team has not decided which facts must always be checked, who owns the final decision, or how a correction is recorded.

Hiring an engineer might improve the integration. It would not, by itself, create that accountability. A caseworker who knows the rules, a supervisor who can design the check, and a technical adviser who can test the system may be a better combination. One person in a small team may hold more than one of those roles. The responsibilities should still be visible. This scene is hypothetical. It is not a reported incident.

Flexible talent does not transfer judgment by itself

The 63% preference for flexible models can be sensible for a small organization if the outside person leaves a task inventory, test cases, data safeguards and an escalation path, and if internal staff keep the decisions. A contractor who does not know the clients, or the informal exceptions that keep a service safe, can also leave a polished framework that nobody can apply after the contract ends. The aim is to buy scarce technical skill for a while and build durable judgment inside. That is an operating suggestion. The Randstad survey did not test it.²

European adoption is a separate caution, and two official pages do not agree on the enterprise figure. The Joint Research Centre publications record for AI and Skills (JRC148616), dated 15 October 2026, by Alice Bertoletti, Alessia De Santo, Yannis Galanakis and Wouter Vergote, says that in 2024-2025 30% of EU workers reported using AI at work, and that by 2025 20% of enterprises had adopted AI technologies. Use ranges from close to 40% among white-collar workers to 6% in elementary occupations, and is concentrated in large firms, digitally intensive sectors and Northern Member States. Exposure is not replacement: the abstract says whether AI augments or displaces workers depends on how central judgement, responsibility and human interaction remain. Employment and vacancy data, it says, show no clear sign of labour-market disruption. The official PDF did not open for this article. The DOI landing page returned fragmented text, so it is not quoted as a clean abstract.⁶

A Directorate-General for Employment news page of 23 October 2026 summarises the same brief. It agrees on around 30% of workers, around 40% of white-collar workers and 6% in elementary jobs. It also says that up to 2025 there is no clear link between AI exposure and unemployment, earnings or working hours in Europe. That sentence is the news page's, not a line this article verified in the PDF. The same page says about 30% of EU enterprises with ten or more employees have adopted AI. That does not match the abstract's 20% of enterprises. This article does not merge the two enterprise numbers, and it does not treat the news page's labour-market sentence as a stronger finding than the abstract's "no clear sign" in employment and vacancy data.⁷

Do not build the next gap by hiring only from engineering networks

The press release says women account for 31% of professionals with emerging AI skills, compared with 36% of those with established tech skills. It then says this share was getting smaller, with Gen Z women accounting for 37% of those with established tech skills but "now only" 32% of those with emerging AI skills. The overview uses a stronger verb: women account for 31% in emerging AI skills, "down from" 36% in legacy tech roles. These are shares inside the candidate-profile data, not a census, and not a published cohort of the same people over time. "Compared with" and "down from" are not the same claim. This article keeps the press comparison and does not adopt "down from" as a measured decline.¹²

If a team recruits only from familiar engineering networks, it can miss people who already do the hard part of governance: supervisors who resolve exceptions, coordinators who document handovers, compliance specialists who challenge weak evidence, and service staff who explain a decision to a client. Domain knowledge is a starting asset. It is not, by itself, proof that the person can test an unreliable output.

The opposite mistake is to rename ordinary communication as a premium AI skill without a demonstration. Ask for a worked example: an ambiguous output, an explanation a colleague can use, and a recorded correction. Keyword lists are a weak substitute. For how generated applications can hide that evidence, see why the AI application flood conceals real skills. The employment gap in AI-exposed US occupations, widest without a degree, is a different study. See why that gender gap is widest without a degree. The two-speed market for AI skills is a third picture. See why the skills boom is two job markets.

A 30-day test is a suggestion, not a validated protocol

A small organization does not need a new department to start. The sequence below is an editorial proposal. Randstad did not test it, and it is not a research finding.

In the first week, pick one recurring workflow. Write down the task, the data, the people affected and the cost of a mistake. Say what the system will do, what stays human, and where the process must stop.

In the second week, collect a small set of real examples, including known exceptions. Agree the expected outcome before running the tool. Record errors by severity. A rare mistake about eligibility or confidential information can matter more than many formatting slips.

In the third week, rehearse the handoff with the people who do the work, not only the person who bought the software. Can they explain why an output is questionable? Can a manager resolve a disagreement? Is there a record? Protect paid time for the practice.

In the fourth week, look at four results: time saved after the checking and the corrections; the rate and seriousness of errors; how many staff can handle an exception without the original buyer; and the effect on workload, confidence and service quality. If the results are weak, narrow the task or pause. The ability to stop is part of governance.

The legal line is narrower than a general warning

This article is not legal advice. For organizations in the European Union, a few provisions describe duties that a "hire more communicators" slogan does not cover.

Article 4 of the EU AI Act, in the text the Commission service desk shows as amended, says providers and deployers shall take measures to support the AI literacy of their staff and other people who operate AI systems on their behalf. They must take account of technical knowledge, experience, education and training, the context of use, and the people or groups the systems are used on. The same text says this obligation does not require anyone to guarantee a specific level of AI literacy for any individual. The Commission and the Member States shall support those efforts, in particular for SMEs. The service desk says the wording reflects the EUR-Lex consolidated text as at 27 July 2026, including Digital Omnibus amendments. The summary on that page is not legally binding.⁸

Annex III lists, among others, AI systems intended for recruitment or selection of natural persons, including targeted job advertisements, analysing and filtering applications, and evaluating candidates. It also lists systems intended to decide terms of a work relationship, promotion or termination, to allocate tasks based on individual behaviour or personal traits, or to monitor and evaluate performance. A further point covers systems intended to be used by public authorities, or on their behalf, to evaluate eligibility for essential public assistance benefits and services. A private charity's summary tool is not, by that description alone, one of those systems. Classification depends on what the technology is intended to do. The service desk says Annex III rules apply from 2 December 2027. Prohibited practices, some transparency duties and the rules for general-purpose AI models are enforceable from 2 August 2026.⁹¹⁰

Where a system is high-risk, Article 26(2) says the deployer shall assign human oversight to natural persons who have the necessary competence, training and authority, and the necessary support. Article 26(7) says that before a high-risk system is used at the workplace, employers shall inform workers' representatives and the affected workers. Those duties attach to the system and the moment of use. They are not satisfied by a headcount of "governance roles" on a careers page.¹¹

Article 5(1)(f) prohibits placing on the market, putting into service for this purpose, or using AI systems to infer emotions of a natural person in the workplace and in education institutions, except where the use is intended for medical or safety reasons. A skill of judgment is not an emotion score.¹²

GDPR Article 22(1) gives a person the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects or similarly significantly affects them. Paragraph 2 sets out exceptions, including contract, a qualifying law, and explicit consent. For the contract and consent cases, paragraph 3 requires safeguards, at least the right to obtain human intervention, to express a point of view and to contest the decision. In case C-634/21 the Court of Justice read recital 71 as covering, for example, e-recruiting practices without human intervention, and it set three cumulative conditions: a decision, based solely on automated processing, producing legal or similarly significant effects. A checklist that nobody is allowed to overrule is not automatically outside that provision. Nor is every drafting tool inside it.¹³

Questions leaders actually ask

Does 61% mean coding skills are now easy to hire?

No. The 61% is a comparison reported by senior decision-makers in this survey. The same release says entry-level tech jobs requiring AI skills are up fivefold since 2025, and that 86% of the surveyed organizations are expanding technology headcount. The wording of the 61% is not identical on every page. None of those lines is a count of unfilled coder jobs.¹²

Does 108% against 69% mean 39% of governance jobs are empty?

No. The press says demand doubled (+108%) while talent supply grew by 69%. The figures come from different data. Subtracting them does not produce a vacancy rate.²

Should a small employer copy the 63% and hire a contractor first?

Not as a rule. 63% of these respondents named flexible talent as their first choice. Randstad sells that model. A short engagement can help if knowledge stays inside. It can also leave a framework the team cannot run. The survey did not measure which of those outcomes happened.²

How hilighter can help

hilighter does not recruit, score candidates, certify governance skills, or predict shortages. The practical question this evidence leaves open is narrower: which roles still require a person to challenge an unreliable output, who has shown that they can do it, and who is allowed to stop the workflow.

hilighter helps teams map skills by role, compare people with those requirements in Skills & Talent Insights, record proof of capability in Skills Evidence & Recognition and set personalized growth paths. Its AI Job Impact view is indicative. It is not a prediction that a role will disappear. A free plan covers up to three job roles, with unlimited employees.

Sources

1. Randstad, Tech talent in transition: building the AI-ready workforce, report overview, consulted 2 November 2026. Figures and wordings used here: 1,100 senior leaders, 11 markets, over 20 million job postings; 86% expanding tech headcount; entry-level postings requiring AI skills up more than fivefold since 2025; 61% on human capabilities versus technical coding skills, and the FAQ wording on communication, critical thinking and stakeholder management versus pure technical coders; demand for risk governance doubled (+108%); 63% flexible talent; women 31% of emerging AI skills, "down from" 36% in legacy tech roles.
2. Randstad, 86% of employers are growing tech teams as AI reshapes demand, press release, 29 October 2026. Methodology: first half of August 2026, 11 named markets, 1,100 interviews, 100 per market, senior business and technology decision-makers, Evalueserve for sampling. Secondary research by Randstad Market Intelligence, August 2026, over 20 million job postings and approximately 100 million candidate profiles. Figures used here include the 61% wording variants, fivefold claims, 86%, +108% demand and 69% supply, 63%, 92% and 42% on digital talent marketplaces, gender shares 31% versus 36% and Gen Z 37% versus 32%, and the CEO quotation.
3. OECD, Building an AI-ready public workforce, 19 January 2026. Three workforce groups and the skill lists quoted in this article. The page cites the Alan Turing Institute (2024) and the OECD public-sector digital talent framework (2021). This is public-administration guidance.
4. OECD, AI and skills: What we know so far, 5 June 2026. Opening claim on skills versus technology; less than 1% need advanced AI skills (Green and Lamby, 2023); digital, data, managerial and human skills; empathy finding of 20% versus 12% in Germany, France, Italy and Spain (Milanez, Lemmens and Ruggiu, 2025), with the brief's caution that it is too early for firm conclusions. The brief says its most recent data date from late 2024.
5. International Labour Organization and partners, Changing landscape of skills in the age of AI, publication page dated 13 August 2026. This article uses the page summary: higher-order cognitive and socioemotional skills, general digital and data science skills, and specialist AI jobs described as currently small and niche but growing rapidly. The full PDF was not re-read page by page.
6. European Commission, Joint Research Centre, Alice Bertoletti, Alessia De Santo, Yannis Galanakis and Wouter Vergote, AI and Skills, JRC148616, publications record dated 15 October 2026, DOI 10.2760/7588639. The PDF was not opened. The DOI page text was fragmented on 2 November 2026 and is not quoted. The abstract used here is the publications record: 30% of EU workers in 2024-2025, 20% of enterprises by 2025, close to 40% of white-collar workers, 6% in elementary occupations, concentration in large firms, digitally intensive sectors and Northern Member States, judgement and responsibility as the hinge between augmentation and displacement, and no clear sign of labour-market disruption in employment and vacancy data.
7. European Commission, Directorate-General for Employment, Social Affairs and Inclusion, AI and skills in Europe - Closing the adoption gap, 23 October 2026. Agrees on around 30% of workers and the occupation contrast. States about 30% of EU enterprises with ten or more employees, which does not match the JRC abstract's 20%. States that up to 2025 there is no clear link between AI exposure and unemployment, earnings or working hours. That sentence was not verified in the JRC PDF.
8. European Commission, AI Act Service Desk, Article 4: AI literacy. The page says the text reflects the EUR-Lex consolidated version as at 27 July 2026, with Digital Omnibus amendments marked, and that the summary is not legally binding.
9. European Commission, AI Act Service Desk, Annex III, point 4 (employment, workers' management and access to self-employment) and point 5(a) (eligibility for essential public assistance).
10. European Commission, AI Act Service Desk, When does enforcement start? (Annex III rules from 2 December 2027; prohibited practices, certain transparency duties and general-purpose AI model rules enforceable from 2 August 2026).
11. European Commission, AI Act Service Desk, Article 26, paragraphs 2 and 7, on the same consolidated text.
12. European Commission, AI Act Service Desk, Article 5, paragraph 1(f), emotion inference in the workplace and in education institutions, with the medical or safety exception.
13. Regulation (EU) 2016/679, Article 22. Court of Justice, case C-634/21, on the three conditions and the recital 71 example of e-recruiting without human intervention.

Sources last reviewed: 2 November 2026. Randstad figures are the publisher's survey and posting analysis. The 61% wording is not identical across the overview and the press release. 108% and 69% are not a vacancy count. The gender lines "compared with" and "down from" are not treated as one measured decline. The JRC PDF was not opened. The enterprise figure on the 23 October news page (about 30% of enterprises with ten or more employees) conflicts with the publications-record abstract (20% of enterprises). The ILO full report was not re-read page by page. EU dates follow the Commission service desk and are not legal advice. Neither an AI-detector score nor the absence of Google penalties can be guaranteed.